基于传感融合的智能垃圾清理车平台设计研究  

Research on the Design of Intelligent Garbage Collector Platform Based on Sensor Fusion

作　　者：李颖捷 陈思洁 时启硕 陈鹏[2] LI Yingjie;CHEN Sijie;SHI Qishuo;CHEN Peng(School of Physical Science and Technology,Southwest Jiaotong University,Chengdu,610031,China;School of Mechanical Engineering,Southwest Jiaotong University,Chengdu,610031,China)

机构地区：[1]西南交通大学物理科学与技术学院,四川成都,610031 [2]西南交通大学机械工程学院,四川成都,610031

出　　处：《系统仿真技术》2019年第3期202-207,共6页System Simulation Technology

基　　金：中央高校基本科研业务费专项资金资助项目(2682017CX036)

摘　　要：针对嵌入式处理器GPU运算能力弱的问题,以Kinect摄像头作为视觉传感器,提出了一种适用于移动机器人的结合ORB(Oriented FAST and Rotated BRIEF)特征和视觉词典的视觉惯性SLAM(Simultaneous Localization And Mapping)算法。该算法首先从彩色图像中提取出ORB特征,然后采用快速近似最近邻(FLANN)算法进行特征匹配,再用PnP(Perspective-n-Point)求解出相机的初始位姿,并用非线性方法对相机的位姿进行优化,而后将上述视觉位姿信息与惯性进行组合实现联合位姿估计。在后端,采用基于BoW(Bag-of-Words)的回环检测来解决位姿估计随时间漂移的问题。实验结果表明,该算法能够很好地为移动机器人提供自主导航与定位。Aiming at the problem of weak computing power of embedded processor GPU,with Kinect camera as visual sensor,a visual inertia SLAM(Simultaneous Localization And Mapping) is proposed for the combination of ORB(Oriented FAST and Rotated BRIEF) features and visual dictionary of mobile robot’s algorithm.Firstly,the algorithm extracts the ORB features from the color image,and then it uses the Fast Library for Approximate Nearest Neighbor(FLANN) algorithm to perform feature matching.Finally,PnP(Perspective-n-Point) is applied to solve the initial pose of the camera,and nonlinear methods are introduced to optimize the pose of the camera.Combining the above visual pose information with inertia,the joint pose estimation is implemented.On the back end,BoW(Bag-of-Words) based loop detection is conducted to solve the problem of position estimation drifting with time.The experimental results show that the algorithm can provide autonomous navigation and positioning for mobile robots.

关 键 词：同时定位与地图构建 ORB特征 惯性测量单元(IMU) 视觉词典 

分 类 号：TP249[自动化与计算机技术—检测技术与自动化装置]